Abstract
INTRODUCTION
Dementia commonly complicates Parkinson disease (PD) and is associated with increased mortality [1, 2]. While cortical Lewy body pathology associates most strongly with PD dementia (PDD), Alzheimer disease (AD) pathology also associates with PDD. Indeed, patients with PDD have greater AD pathology than PD patients without dementia [3, 4]. Numerous studies have found that the APOEɛ4 allele, the major genetic risk factor for AD, is associated with cognitive impairment in PD. [5, 6] Meta-analysis of genome-wide association studies in AD have implicated multiple single nucleotide polymorphisms (SNPs) and associated genes other than APOE, notably BIN1, CLU, ABCA7, CR1, PICALM, MS4A6A, CD33, MS4A4E, and CD2AP [7]. The role of these AD risk genes in cognitive impairment in PD remains incompletely evaluated. The objective of this study was to test alleles associated with risk of AD for association with cognitive impairment in PD in two separate PD cohorts.
METHODS
Datasets and genotyping
For this study, we accessed 2 datasets from the Database of Genotypes and Phenotypes (dbGaP) of the National Center for Biotechnology Information. The Institutional Review Board for Health Sciences Research at the University of Virginia and the dbGaP data access committee approved the use of these data for this study. The. NeuroGenetics Research Consortium collected DNA samples and clinical information from approximately 2000 PD subjects and 2000 non-PD controls that comprise the first dataset, “Genome-Wide Association Study of Parkinson Disease: Genes and Environment” (dbGaP Study Accession: Phs000196.v2.p1) (NGRC). Genotyping of 1,051,295 SNPs was performed using the HumanOmni1-Quad_v1-0_B platform (Illumina, Inc.). In this population, 1159 PD subjects had mini-mental state exam (MMSE) scores. The second dataset, “CIDR: Genome Wide Association Study in Familial Parkinson Disease”, was formed from two genetic studies of PD, PROGENI and GenePD (dbGaP Study Accession: Phs000126.v1.p1) (PROGENI/GenePD). For this cohort, DNA samples were collected for approximately 1000 PD subjects with 2 or more family members with PD and genotyped using the HumanCNV370v1 platform (Illumina, Inc.), which includes 370,404 SNPs. In this dataset 529 PD subjects had MMSE scores. Subjects in both datasets were white and non-Hispanic. The NGRC and PROGENI/GenePD cohorts were the only datasets available in dbGaP that included PD subjects with MMSE scores. Prior to quality control procedures of genetic data, PD subjects from these two datasets were only excluded if they did not have an MMSE score.
Quality control
Prior to analysis, we performed quality control of genetic data with assessment of sex mismatches, informative missingness, heterozygosity, Hardy-Weinberg equilibrium, and cryptic relatedness, and we also applied minor allele frequency and missingness thresholds for individual subjects and individual SNPs. Samples identified as problematic were excluded from further analysis. Application of missingness thresholds was the most common reason for exclusion. After quality control procedures, 997 and 471 PD subjects with MMSE scores from the NGRC and the PROGENI/GenePD cohorts, respectively, were available for analysis.
Selection of Candidate SNPs
The AlzGene database was created as a publicly available, comprehensive catalog of all genetic association studies in Alzheimer disease (http://www.alzgene.org). (Bertram et al., 2007) On December 1, 2013, the database included 1395 AD genome-wide association studies (GWAS). Use of the HuGENet interim criteria for the cumulative assessment of genetic associations to rank genes associated with AD identified 9 AD risk genes other than APOE, each represented by one or more significant SNPs. With the number of significant SNPs in parentheses, the genes are BIN1 (2), CLU (5), ABCA7 (1), CR1 (4), PICALM (5), MS4A6A (1), CD33 (1), MS4A4E (1), and CD2AP (1). We selected for investigation SNPs reaching significance in meta-analyses for each of these genes.
Statistical analysis
After extracting available SNPs and removing SNPs that failed quality control, 6 SNPs were available for analysis in the NGRC cohort and 10 SNPs were available for analysis in the PROGENI/GenePD cohort. Logistic regression and ordinal regression models, adjusted for sex, age at MMSE, and duration of PD, were constructed to assess the association between selected SNPs and MMSE score in each dataset separately. In logistic regression models, MMSE was dichotomized at 24 (< 24, ≥24). This cut-point corresponds to the traditional cut-point for dementia and offers optimal diagnostic accuracy for identification of PD-D and PD-MCI [8, 9]. Ordinal regression models with MMSE as dependent variable were used because MMSE scores were not normally distributed. For the NGRC cohort, models included covariates for APOE allele status and family history of PD. To evaluate the relationship between APOE allele status and PD in the NGRC cohort, we constructed logistic regression models with dichotomized MMSE as the dependent variable and APOE ɛ4 carrier status as the predictor of interest. APOE allele status was not available in the PROGENI/GenePD dataset. We used 3 principal components to adjust for population structure. As the primary analyses, we examined each individual allele in relation to MMSE in logistic regression models evaluating the additive effect of allele dosage. In the PROGENI/GenePD cohort, we had greater than 80% power to detect an OR of 2.0 for risk of dementia associated with the minor allele, and in the NGRC cohort, we had greater than 80% power to detect an OR of 1.6 for risk of dementia associated with the minor allele. We assessed dominant and recessive models as exploratory analyses. These did not reveal additional significant SNPs. We adjusted p-values for multiple comparisons using false discovery rate (FDR) and set significance at p < 0.05. We performed age-stratified analyses (> 70, ≤70) for the 2 SNPs associated with MMSE using unadjusted p-value < 0.05. We performed statistical analyses using R (R Foundation for Statistical Computing, Vienna, Austria) combined with Plink v1.07 (Cambridge, MA).
RESULTS
Table 1 contains demographic and clinical characteristics of both datasets. For PD subjects with MMSE scores in the NGRC dataset (n = 997), the mean was 27.5 (SD = 3.3), the median was 28 (interquartile range = 25.5, 30) and 13.4% had MMSE < 24. For PD subjects with MMSE scores in the PROGENI/GenePD dataset (n = 471), the mean was 27.5 (SD = 3.3), the median was 29 (interquartile range = 27, 30) and 9.0% had MMSE < 24. In logistic regression models adjusted for 3 principal components, sex, age at MMSE, and duration of PD at time of MMSE, carriers of APOE ɛ4 allele had a greater, but non-significant, risk of MMSE < 24 compared to MMSE≥24 (OR = 1.6; p = 0.09). In similarly adjusted logistic regression models, carriers of APOE ɛ4 allele had a significantly greater risk of MMSE < 24 compared to MMSE > 28 (OR = 2.2; p = 0.008).
We found two nominal associations in unadjusted analyses, but neither association remained after adjustment for multiple comparisons. Adjustment for APOE allele status and family history of PD in the NGRC cohort did not significantly alter the results. A SNP in BIN1 (rs744373; chr2.GRCh38.p2:G.127137039A > G) was associated with cognitive impairment (MMSE < 24) (unadjusted p-value = 0.036) in the NGRC cohort (Table 2). Prior to adjustment for multiple comparisons, the BIN1 SNP rs744373 was associated with MMSE < 24 in the NGRC cohort in only those > 70 years old (unadjusted p-value = 0.048; adjusted p-value = 0.19, n = 442).
A SNP for PICALM (rs3851179; chr2.GRCh38.p2:G.86157598T>C) was associated with cognitive impairment (MMSE < 24, unadjusted p-value = 0.043; and worse MMSE score, unadjusted p-value = 0.021) in the PROGENI/GenePD cohort (Table 3). The PICALM SNP rs3851179 was associated with cognitive impairment (MMSE < 24) in PD subjects > 70 years old (OR = 2.3; adjusted p-value = 0.017; n = 250) but not in PD subjects≤70 years old (adjusted p-value = 0.65, n = 221). The SNP rs3851170 was not significantly associated with MMSE score in the NGRC cohort.
DISCUSSION
Using MMSE score as a global cognitive measure, none of the SNPs associated with AD assessed in this study were significantly associated with cognitive impairment in PD. However, analyses stratified by subject age indicate that older age may be an important factor influencing the effect of AD risk SNPs on cognition in PD. PICALM SNP rs3851179 was associated with cognitive impairment in subjects older than 70 years old but not in subjects less than 70 years old in the PROGENI/GenePD cohort. In the NGRC cohort, the BIN1 SNP rs744373 was associated with MMSE score in those older than 70 years old, but this was no longer associated after adjustment for multiple comparisons using FDR.
Prior studies have suggested a greater contribution of AD pathology to dementia in PD patients with older onset [10]. Our results support this observation. Of the 5 SNPs associated with AD at the PICALM locus, rs3851179 is the one most significantly associated with AD. The exact role of PICALM in AD pathogenesis is unknown, but it appears involved in Aβ clearance and co-localizes with tau in mature neurofibrillary tangles [11]. Prior studies reported no association between the PICALM SNP rs3851179 and PD risk [12–14]. The association between the PICALM SNP rs3851179 and the risk of cognitive impairment in PD has not been previously evaluated. The direction of the association previously reported for AD was the opposite of what we found in PD subjects older than 70 [15, 16]. This indicates either our finding is a false positive or perhaps the mechanism by which the PICALM locus contributes to dementia in PD is different from AD. Given our modest sample size, the p-value of 0.06 for the association of BIN1 SNP rs744373 and MMSE in PD patients older than 70 warrants further investigation.
The fact all subjects in the PROGENI/GenePD cohort had a family history of PD may have contributed to the different findings between the NGRC and PROGENI/GenePD cohorts. Genetic risk factors for cognitive impairment, and AD risk factors specifically, may have differential effects in those with and without a family history of PD. The inability to adjust for APOE allele status in the PROGENI/GenePD cohort is a limitation; however, adjustment for APOE allele status in the NGRC cohort did not significantly alter the results of regression analyses. MMSE score provides an imperfect measure of cognition and constitutes a limitation of this study. Multiple studies have demonstrated the superior sensitivity and specificity of the Montreal Cognitive Assessment as a screening cognitive instrument in PD [8]. More refined measures of cognitive functioning would increase precision, which could increase the power to detect genetic differences between groups. Our finding that APOE allele status was significantly associated with MMSE < 24 compared to MMSE > 28 but not MMSE > 24 underscores the relative insensitivity of the MMSE to determine cognitive status. The availability of MMSE scores at only one time-point represents another limitation of this study. Multiple data points in a longitudinally followed cohort would increase precision in defining groups of PD patients and would also allow study of the effect of risk alleles on rate of cognitive decline. Lastly, targeted sequencing rather than reliance on GWAS data with incomplete coverage would allow more complete assessment of whether SNPs and genes associated with AD are also important in the development of cognitive impairment in PD. Our data indicate that, of the AD risk genes identified, the PICALM SNP rs3851179 may contribute to the development of cognitive impairment in older individuals with PD. It is important that future studies include more discriminative cognitive testing and consider the interaction of age and genetic risk factors in the development of cognitive impairment in Parkinson disease.
CONFLICT OF INTEREST
Dr. Barrett, Dr. Koeppel, and Dr. Turner have no conflicts of interest to disclose. Dr. Worrall is an associate editor at the journal Neurology.
Footnotes
ACKNOWLEDGMENTS
This work was supported by Award No. 14–1 from the Commonwealth of Virginia’s Alzheimer’s and Related Diseases Research Award Fund, administered by the Virginia Center on Aging, School of Allied Health Professions, Virginia Commonwealth University.
This work utilized data from the NINDS dbGaP database (
), including the dataset CIDR: NGRC PARKINSON’S DISEASE STUDY (dbGaP Study Accession: Phs000196.v2.p1). The NeuroGenetics Research Consortium (NGRC) was funded by the NIH and led by PI Haydeh Payami (R01 NS36960). Contributing investigators were Drs. Stewart Factor (Emory University), John Nutt (Oregon Health & Sciences University), Cyrus Zabetian (University of Washington and Puget Sound Veterans Medical Center), Eric Molho (Albany Medical College), and Donald Higgins (Albany Veterans Medical Center). NGRC’s molecular and statistical genetics laboratories are at New York State Department of Health (Haydeh Payami) and Puget Sound Veterans Medical Center (Cyrus Zabetian). This work also used the dataset CIDR: Genome Wide Association Study in Familial Parkinson Disease (dbGaP Study Accession: Phs000126.v1.p1) from the NINDS DbGaP database. This dataset was composed of two studies, PROGENI and GenePD. Funding support for PROGENI (PI: Tatiana Foroud; R01NS037167) and GenePD (PI: Richard Myers; R01NS036711) were provided by the NIH and the genotyping of samples was provided by the National Institute of Neurological Disorders and Stroke (NINDS).